Examining methods to estimate static body sway from the Kinect V2.0 skeletal data

Abstract

Static body sway is a clinically relevant activity parameter, used to assess postural balance, across a wide spectrum of patient populations. We have examined static body sway using two different segmental total body center of mass (TBCM) estimation methods, the Generator of Body Data III (GEBOD) and Winter's method, using Microsoft Kinect skeletal data. Twenty subjects were recruited through an IRB study and asked to perform three trials of single leg stance with their eyes closed, with positioning based on the Balance Error Scoring System. A force plate system was used to estimate the ground truth data for comparison. Results show that both GEBOD and Winter's method performed similar in estimating anterior-posterior (AP) and medio-lateral (ML) body sway. The results also show highly correlated measurements by the two TBCM estimation methods when compared with the force plate system (mean RMSE value of 10.18 mm square in AP and 8.00 mm square in ML direction). Ordinary Least Square (OLS) linear regressions were performed to improve body sway results obtained from the two methods. Improved sway range values obtained from the simple regression method was able to reduce the estimation errors by 50% (~ 10 mm in both AP and ML body sway). The two static body sway estimation methods were found reliable for obtaining body sway. Thus, the inexpensive, portable Kinect V2.0 can be used for clinical measurements.